Short-Term Load Forecasting Based on LightGBM Parallel Ensemble Method
High-accuracy short-term load forecasting can effectively guide the start-up and shutdown plans of coal generating units and reduce the waste of coal resources. To improve the short-term load forecasting accuracy, a LightGBM parallel Ensemble paradigm-based short-term load forecasting method for electric power is proposed. Firstly, the relevant input features of the load to be predicted are screened using the Spearman coefficients; then LightGBM, an efficient serial Ensemble method, is selected as the base learner, its parameters are optimized using the Ant lion optimizer, and cross-validation is used to ensure the diversity of different LightGBM base learners. Finally, the Bagging ensemble method is used as a parallel ensemble paradigm to achieve forecasting with the optimal weighting method combined with the base learners. Experimental analysis using Australian electricity consumption data as an arithmetic example shows that the LightGBM-based parallel ensemble method combines the advantages of serial and parallel ensemble methods to simultaneously reduce forecast bias and variance and improve the quality and speed of short-term electricity load forecasting.